TinyTorch from Harvard/ETH: build your own ML framework in pure Python, no GPU needed
lmoroney · x · 2026-10-12
TinyTorch is a free open-source curriculum born from Harvard's CS 249r Machine Learning Systems project, now maintained from ETH Zurich. You build a working ML framework in pure Python across 20 modules — tensors through transformers — using PyTorch's own API, on a 4 GB RAM laptop with no GPU.
- Module 01 has you compute memory footprints before writing matmul; later you measure why Adam needs 3x the optimizer memory of SGD
- Slowness is the point: your Conv2d taking 97 seconds vs PyTorch's 10 ms makes vectorization stick forever
- Team is candid: preview stage, CPU only, learning outcomes unmeasured
- Tip: pick the subsystem that keeps biting you (autograd, attention) and build that module this week
Related event: Harvard and ETH Launch TinyTorch, a Pure-Python ML Framework(2 posts)→
More from coding & agent
- What if every query had its own harness? — amankhan · 2026-10-12
- MCP Server for Images: Compress, Convert, and Optimize a Whole Site's Images Safely — Able-Place · 2026-10-12
- Frontier models now handle day-long tasks; the bottleneck is your 'bandwidth', says new long-form analysis — dotey · 2026-10-12
- Developer has AI agents build profiling tools and benchmarks to find system bottlenecks — tristanbob · 2026-10-12
- Unlinked: open-source Rust project clones most of Simulink, with MATLAB-to-Rust compiler — rickasaurus · 2026-10-12
- Agent uses Computer Use to reverse-engineer CapCut, batch-downloads 2,291 subtitle templates — dotey · 2026-10-12